NIMROD TEAM MEETING
Monday, Feb 2
SAIC SAN DIEGO
Monday 2 Feb.
1. STATUS OF CODE
- Advection
Predictor-corrector. Same status as at APS. Carl is looking
into upwinding with finite elements. Truncation errors in
Ohm's law may lead to spurious numerical resistivity
and reconnection.
- Two Fluid
Nothing new since APS. This is not a priority until we
pass the GA acceptance test.
- Pressure
Nothing new since APS.
- Vacuum/Separatrix
Carl and Tom have completed triangulation of the vacuum region.
It is ready for Dalton and Alexander to start the
vacuum/plasma problem.
This can only run with EFIT equilibria, not TOQ.
- Organization
Code has been reorganized and cleaned up since APS. Take
new draw.in files from the package.
Management/integrand routine interface has been cleaned up.
2, CODE PERFORMANCE
Linear Solver:
iterative solve takes 75 to 80 percent of the time in NIMROD.
Alfonso has compared ISIS with the NIMROD solver on a NIMROD
matrix. The problem was a square grid with wave propagation.
ISIS seems to be about a factor of 2 faster on a big 32 processor
problem (factor of 3 for single processor). Care should be
taken that the tolerances and norms are the same for the 2 cases.
Steve has been looking at the same type of comparison with AZTEC.
The problem was a circular grid with a central tblock. Got a
factor of 2 faster single processor performance.
We need to test the hardest or most relevant case with both
of these packages.
NIMROD F90 package has:
- Block Jacobi
- Block direct (different block)
- Block ILU_0,1
-1D direct solve along a grid line. (new since APS, in
the readme and input.f files) (these are really "line Jacobi")
DECISION: Alfonso and Steve will continue to test ISIS and AZTEC.
Tests will be done on the same problem. This problem will be
typical of a hard NIMROD physics problem, including the degenerate
rblock and tblocks. Alfonso will use the same driver as Steve.
They should verify the speedup we can expect from using one of
these packages. Then we can make the decision on whether to
incorporate thee packages into NIMROD.
Single Processor Performance: nothing new. Alan thinks everything
is "fine". We'll revisit this when the solver gets optimized.
Parallel FFTs: Steve and Carl have looked at parallelizing
over both grid blocks and Fourier modes. This way we could
get parallelization over the third dimension. Concensus:
Steve and Carl should continue to pursue this, but Steve
should make the linear solver his top priority.
The Dreaded Jaggies: These occur in a linear tearing mode
problem, not near the singular surface but near the boundary.
Alan has identified the cause as imperfect cancellation between
two terms in the linearized Lorentz force. The cure is to
re-organize the force into conservation form. This allows
the finite element formulation of the terms in the Lorentz
force to balance. This also introduces a numerical parallel
force proportional to divB. This can be cured by a well
known method due to Barnes and Brackbill. This has not yet
been verified in practice.
Use of T3E Allocation: Everyone will be reasonable. We'll
try to get more time through Mike Crisp.
Dalton will try to remember to send out status setcub reports
every 2 weeks. Everyone will learn to use setcub itself.
3. VALIDATION PLAN
Required Linear Benchmarks:
GATO ideal kink: Done except for poloidal variation
in vphi. This seems to be a temporal convergence issue.
This is complete.
Soloviev:
Convergence studies are complete. Good agreement with
Berger, et al.
Toroidal Resistive Modes:
Comparison with FAR for S=10**7. Scott is running
these. He had some normalization problems with the case.
Rumors are that he has gotten some good results in the last
few days. Tom will check his e-mail for results.
Galkin case:
Carl and Alfonso have tried to run this. They get
too fast growth rate, but may be converging on eigenfunction.
This is still being investigated.
Whistler waves and 2-fluid effects:
Effect on growing modes in DEBS is unacceptably large.
This is the same algorithm as in NIMROD, except that NIMROD doesn't
have the electron pressure yet. We probably need an antisymmetric
solver to get this right.
Nonlinear cases:
DIIID: n=0 and 1 done, with no advection. Needs to be
re-run with more modes and advection.
Cylindrical tearing: Nonlinear saturation comparison with
DEBS has been done.
Finite (R/a) RFP: Theta=1.6 case reverses and sustains
with nonlinear dynamo (n=0,10), R/a=2. Probably need more modes
(n=20??).
Ballooning (Linear): Eigenfunction looks good (n=10).
Growth rate vs. beta has been calculated. Results are consistent
from DCON, and with growth rate estimates.
GA Acceptance Criteria:
We'll see.
Collaboration with Keldysh: Sergei Galkin gave a presentation
on the work done at Keldysh in support of NIMROD. He gave us a
proposal for the next years work, and we will take it under
advisement for discussion with DOE, who eventually funds the
work. (Thus was given on Tuesday morning, 2/3.)
4, STATUS OF GUI
Tom has kept it up to date. The customers (GA) will
give us the best feedback. Remote operation on the T3E has not yet been
defined by the customer. We'll need to get feedback on this from the GA
beta test.
Tuesday, Feb. 3
5. GRAPHICS AND VISUALIZATION
Alan talked about ANALYZE and NIMPLOT. ANALYZE is
an automated, non-interactive version of NIMPLOT. ANALYZE
will be able to generate all types of output in all types of
formats. Analyze needs to be generated to handle multiple
Fourier modes. Poincare plots will be added to both ANALYZE
and NIMPLOT.
Alan is responsible for ANALYZE. Carl is responsible
for NIMPLOT.
XDRAW is now fully functional on the T3E.
6. DEVELOPMENT
Gridding of Vacuum Region:
The code is ready for implementation of the nonlinear
vacuum-plasma interface problem. Tblocks and gridding of the
vacuum region have been implemented.
Alan will begin work on extending the concept of flux
coordinates into the vacuum region so that we can use rblocks
to the maximum extent possible.
Carl and Dalton will implement the vacuum region algorithm.
Grid Packing near Rational Surfaces:
Alan is starting to think about this problem. He won't
start on this until he cures the dreaded jaggies.
Resistive Wall Formulation:
Alexander gave a presentation and a write up about a
possible resistive wall formulation. It amounts to a time
dependent boundary condition. It would be easier to implement
in the wall resistivity is small so that it could be done
explicitly. Everyone seems to be happy with this formulation.
Alan thinks we may be able to adapt Morrl Chance's code for
this task, but not everyone agrees. Alexander will perform this
task. He will set up a collaboration with Morrel.
Neoclassical Stuff: Tom is working on a self-consistent formulation
of the viscous stress tensor.
7. DOCUMENTATION
Dalton has completed 15 pages of summary documentation.
He and Alfonso will find the best way to get it onto the web site.
We are reminded that SAIC is responsible for the web site,
so Alfonso will fix the existing problems with the html pointers.
Alan is moving the nimrod e-mail reflector to
nimrod.lanl.gov.
8. MANAGEMENT
CVS:
Carl is librarian. He is the only one who can update
the master copy.
On nimrod you need: ==> setenv CVSROOT /usr/local/cvs
cvs so nimrod ==> check out nimrod
cvs diff ==> gives listing of changes
cvs update [filename] ==> flags conflicts, update if none
cvs export -r [HEAD] [nimrod2_1] nimrod==> get source files
9. MEETINGS/PAPERS
Sherwood:
Abstracts due Feb. 11
1. Tearing modes: Alan, Scott??
2. Ballooning modes/Nonlinear internal kink: Tom
3. Status/Review: Carl
4. Soloviev: Alfonso,Dalton
Dalton will not attend Sherwood.
Alan will run and organize the team meeting at Sherwood.
EPS:
Alan is giving an invited talk at EPS. in June/July
IAEA:
Yokohama in 10/98. Abstracts due end of March. We intend to
submit an abstract and we'll wait and see what we have by the
middle of March.
10. OTHER BUSINESS
Budgets, etc.
Advisory board: ??? decision postponed.
12. TIME AND DATE FOR NEXT:
Meeting: Sherwood
Teleconference: Thursday February 19, 12-2pm EST, 9am-11am PST.